How it works

Exhaustive, by construction.

Gage Law answers only from what it has read, and it reads everything a question touches: the whole record, the full opinions, the statute as it stood on the day, and the arguments against you. Then it proves the paper before anyone else sees it.

Four commitments, enforced in code.

None of these is a policy someone has to remember. Each is a check that runs on every question and every paper, and a check that cannot run blocks the work rather than letting it through.

  • Read, never remembered

    An answer is built only from text the system has read in this session: the opinion in full, the statute in the version that governed, the record itself. It draws on 9,427,210 opinions joined by 77,481,570 citation links, and nothing is supplied from a model’s memory.

  • Every claim chained to its proof

    A fact carries the exhibit that proves it, a quotation carries the text it was read from, and every event lands on one timeline. A claim with nothing behind it is not softened; it is removed.

  • Looks for what defeats the argument

    Retrieval tuned to find support finds support. The system also searches the language courts use when they refuse, and answers every authority the other side cited, so the weakness is found before the other side finds it.

  • Proven before it leaves

    A paper passes 77 stages in 10 lanes before it is called ready, behind 1,003 safeguards. 884 of them have been watched catching the very defect they exist for, on material they were never built from.

From a question to an answer: 10 stages.

Finding cases that agree is the easy half. The work is finding the ones that matter, choosing the smallest set that carries the argument, and checking that every one of them is still good law. Each stage below is independently checkable.

  1. Citation resolution

    A reporter triple — volume, reporter, page — resolves to a cluster, then to the opinion text. A lookup, not a search.

    full reporter index

  2. Lexical retrieval

    Full-text search across the opinion corpus, for the case that uses your words.

    52 indexes

  3. Dense retrieval

    Vector similarity across indexed passages, for the case that means the same thing in different words.

    ~2.3M passages

  4. Fusion

    Reciprocal rank fusion combines the lexical and dense rankings into one, so neither mode can hide the other’s best hit.

    every query

  5. Associative recall

    A holographic reduced-representation triple store returns entity neighbours — what is related, not merely what is similar. This is the authority you would not have known to ask for.

    1,556,170 triples

  6. Rerank

    One canonical holographic-similarity reranker turns recall into precision. One strong reranker, never a weak one stacked on it.

    every query

  7. Authority selection

    A brief does not need every case that agrees with it; it needs the smallest set that covers the argument. That is minimum-redundancy / maximum-relevance, expressed as a QUBO and solved by quantum annealing.

    pool from 9,427,210 opinions

  8. Optimisation cross-check

    The annealer’s answer is checked against an exact brute-force oracle where the problem is small enough to permit one, and against an independent exact solver where it is not.

    every solve

  9. Grounding

    Answer-to-context similarity, plus a fabricated-citation guard: a cite that is not present in the retrieved context is flagged unsupported and cannot be smuggled into an answer.

    every answer

  10. Good law

    A negative-treatment scan over the citation graph asks whether an authority has been overruled or superseded — before it reaches a brief, not after.

    every cite

From a draft to a filed paper: 77 stages.

Before a paper is called ready it is measured against the rule of the court it is going to. A draft that fails a stage is not rendered at all, and the stage that stopped it is named, so the fix is known first.

10 lanes, and what each refuses

  1. Authority and citation · 18 stagesNo case is cited until the opinion has been read in full — not a headnote, not a snippet, not a summary. Each is then checked for what the majority actually held, whether the court reached it, and whether it is still good law.
  2. Argument and craft · 15 stagesThe standard of review, who carries the burden and the relief sought are each stated with the authority that sets them. Every authority the other side cited is answered rather than left silent.
  3. Form, forum and filing · 11 stagesCaption, margins, page limits and the filing packet measured against that court’s own rule — and a refusal to draft into a court whose procedure we do not hold, rather than applying another state’s by default.
  4. Source and provenance · 8 stagesEvery fact traces to the document it came from. A search result, an internal note or an earlier draft of our own is a lead to read, never a source to cite.
  5. Adversarial review · 7 stagesSomeone goes looking for what kills the argument before opposing counsel does, including the authority that cuts against us.
  6. Exhibits and service · 5 stagesExhibits numbered without collision, legible at the size they will be read, and reconciled against what the paper says was served on whom.
  7. Release control · 5 stagesNothing regenerates after signature, and nothing leaves without the approval on the record. A signed paper is frozen.
  8. Parties and privacy · 3 stagesEvery party’s identity and role verified from a primary source — no title inferred from a name — and personal data held out of what gets filed.
  9. Dates and clocks · 3 stagesNo “next Tuesday”. Every deadline computed from the rule that sets it and carrying that rule’s citation, so a date can be checked instead of trusted.
  10. Numbers · 2 stagesEvery figure reconciles against its exhibit, and any amount the court may ask about is on one page before anyone stands up.

The filing gate in detail

The record, taken whole.

Exhaustive is a measured claim. These are figures from real work, not projections.

3,651
pages in a single opposing filing, taken in whole: 19 million characters, every exhibit indexed, every authority pulled for reading
899
pages in one exhibit binder, assembled, paginated and checked as a single instrument
1,945
attorney-hours of reading and drafting, worked through by the engine in six days
989
reporters indexed, so a citation resolves to its opinion by lookup, not by search

64.8 million tokens across 14,363 engine calls, 21–26 September 2026. Converted at 0.75 words per token, 500 words a page, 50 pages an attorney hour, $400 an hour.

The proof is published.

  • The evaluation report

    Invented citations caught, real ones confirmed at the reporter, and held-out defects found, with the runs behind each figure.

  • The research

    9 full-length papers built the same way: no claim without the primary source that proves it.

  • The methodology

    What the system does, what it refuses to do, and where its own limits are stated.

See it on a matter of your own.